Building an AI agent with structured content access: Content Agent API + Vercel AI SDK

Conference organizers live in messaging apps. The content backend is where the data lives. The gap between those two places is where things fall through. So I built a Telegram agent for the (fictive) ContentOps Conf that closes that gap. Organizers just message it from their phones. The agent reads and writes the content backend that runs the whole event. Sessions, speakers, submissions, schedule, all of it. Nobody opens a dashboard. The whole thing is about 100 lines of application code. In this video I walk through the three layers that make it work: Sanity Content Agent for knowledge and permissions Vercel AI SDK for streaming and conversation Chat SDK for platform routing The point isn't the agent. It's that agentic content ops now fits inside the tools your team already uses. Telegram today. Slack, Discord, iMessage with a different adapter. The message handler is platform-agnostic, so you swap the adapter and keep everything else. Full writeup with all the code: https://www.sanity.io/blog/build-a-te... Open-source repo (conference starter): https://github.com/sanity-labs/confer... Docs: Content Agent → https://www.sanity.io/docs/content-agent Content Lake → https://www.sanity.io/docs/content-lake GROQ → https://www.sanity.io/docs/groq Vercel AI SDK → https://sdk.vercel.ai Chat SDK → https://github.com/nichochar/chat A companion post covers the attendee-facing side of this same app (read-only, Agent Context + Claude instead of Content Agent API): https://www.sanity.io/blog/build-a-co... Chapters: 0:00 Live demo: asking the agent about submissions 0:49 The three-layer architecture 1:03 Accepting talks through chat 1:15 Studio tour: the conference content model 1:56 Why chat beats opening the laptop 2:40 Layer 1: Content Agent and access filters 3:28 Layer 2: Vercel AI SDK, streaming, and conversation 4:38 Storing conversations in the Content Lake 5:43 Layer 3: Chat SDK and platform adapters 6:24 Demo: rescheduling a session and drafting an announcement 7:48 Open source, the two-agent architecture, companion post